Abstract: Provided is a method for computer-implemented determination of a data-driven prediction model. The method processes digital input data having input and output variables and being semantically annotated based on a digital semantic representation having a hierarchical tree structure where each tree in the structure represents an input variable of the data, the leaf nodes of the respective tree being the discrete values of the input variable. The method of the embodiment provides a recoding of those discrete values by cutting off hierarchical levels of the respective trees. Based on this recoding, a plurality of data modifications is determined for the input data. Those data modifications are trained by a machine learning method where the trained machine learning method with the highest prediction quality is derived from the trained machine learning methods.

Abstract: Aspects and features of the present disclosure can reconcile node sharing within a production rule network that is fully coded in a compiled language. As an example, the shared, stateless class can represent constraints shared by the alpha node of a rete network. Code can be post processed to create a shared stateless class defined in memory. When the rule engine is executed and the rule network is produced, the shared stateless class can be referenced to evaluate a constraint shared by a node of the production rule network, reducing the number of classes stored in memory. Garbage collection can be used within the shared stateless class, deleting objects from memory structure when no longer used, further improving storage efficiency.

Abstract: Embodiments described herein provide safe policy improvement (SPI) in a batch reinforcement learning framework for a task-oriented dialogue. Specifically, a batch reinforcement learning framework for dialogue policy learning is provided, which improves the performance of the dialogue and learns to shape a reward that reasons the invention behind human response rather than just imitating the human demonstration.

Abstract: The present teaching relates to method, system, medium, and implementations for machine learning. A training sample is first received from a source. A prediction is generated according to the training sample and based on one or more parameters associated with a model. A metric characterizing the prediction is also determined. The prediction and the metric are transmitted to the source to facilitate a determination on whether a ground truth label for the training sample is to be provided. When the ground truth label is received from the source, the one or more parameters of the model are updated based on the prediction and the ground truth label.

Abstract: A convolutional neural network (CNN) processing method includes selecting a survival network in a precision convolutional network based on a result of performing a high speed convolution operation between an input and a kernel using a high speed convolutional network, and performing a precision convolution operation between the input and the kernel using the survival network.

Abstract: Pure machine learning classification approaches can result in a “black box” solution where it is impossible to understand why a classifier reached a decision. This disclosure describes generating new classification rules leveraging machine learning techniques. New rules may have to meet evaluation criteria. Legibility of those rules can be improved for understanding. A machine learning classifier can be created that is used to identify possible candidate classification rules (e.g. from a group of decision trees such as a random forest classifier). Classification rules generated with the assistance of machine learning may allow for identification of transaction fraud or other classifications that a human analyst would be unable to identify. A selection process can identify which possible candidate rules are effective. The legibility of those rules can then be improved so that they can be more easily understood by humans.

Abstract: A conversation management system provides a conversational computing interface that manages verbal exchanges between a computer (e.g., artificial intelligence) and a human operator. In a medical embodiment, conversational input from the computing system is generated based on medical knowledge, uses deep learning algorithms, and/or intelligently tracks the state of a conversation so as to the most relevant data to the user.

Abstract: A conversation management system provides a conversational computing interface that manages verbal exchanges between a computer (e.g., artificial intelligence) and a human operator. In a medical embodiment, conversational input from the computing system is generated based on medical knowledge, uses deep learning algorithms, and/or intelligently tracks the state of a conversation so as to the most relevant data to the user.

Abstract: An apparatus and method are described for reducing the parameter density of a deep neural network (DNN). A layer-wise pruning module to prune a specified set of parameters from each layer of a reference dense neural network model to generate a second neural network model having a relatively higher sparsity rate than the reference neural network model; a retraining module to retrain the second neural network model in accordance with a set of training data to generate a retrained second neural network model; and the retraining module to output the retrained second neural network model as a final neural network model if a target sparsity rate has been reached or to provide the retrained second neural network model to the layer-wise pruning model for additional pruning if the target sparsity rate has not been reached.

Abstract: A medical information processing apparatus according to an embodiment includes a processing circuitry. The processing circuitry is configured: to distribute, to an information processing apparatus provided at each of a plurality of medical institutions, a program for causing a machine learning process to be executed by using medical data held at the medical institution having the information processing apparatus; to receive, from each of the information processing apparatuses, a change amount in a parameter related to the machine learning process, regarding a change caused in conjunction with the execution of the machine learning process; to adjust a value of the parameter on the basis of the received change amount; and to transmit the adjusted parameter to each of the information processing apparatuses to cause the machine learning process to be executed on the basis of the parameter.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network system used to control an agent interacting with an environment. One of the methods includes receiving a current observation; processing the current observation using a proposal neural network to generate a proposal output that defines a proposal probability distribution over a set of possible actions that can be performed by the agent to interact with the environment; sampling (i) one or more actions from the set of possible actions in accordance with the proposal probability distribution and (ii) one or more actions randomly from the set of possible actions; processing the current observation and each sampled action using a Q neural network to generate a Q value; and selecting an action using the Q values generated by the Q neural network.

Abstract: A distributed inference system includes an end device and a server. The end device generates status information and generates an inference result corresponding to target data based on a first machine learning model. The server creates a second machine learning model based on the status information and a training dataset including the inference result, calculates an accuracy of the inference result, and provides the second machine learning model to the end device based on the accuracy.

Abstract: An output time-series of a cell of a neural network is captured. A subset of a set of data points of the output time-series is consolidated into a singular data point. The singular data point is fitted in a data representation to form a quantified aggregated data point. The quantified aggregated data point is included in an intermediate time-series. Using the intermediate time-series as an input at an intermediate layer of the neural network, an anonymized output time-series is produced from the neural network.

Abstract: The present teaching relates to method, system, medium, and implementations for machine learning. A check is performed on a level of available bidding currency for bidding a training sample that is used to train a model via machine learning. A bid in an amount within the available bidding currency is sent, to a source of the training sample, for the training sample. The training sample is received from the source when the bid is successful. A prediction is then generated in accordance with the training sample based on one or more parameters associated with the model and is sent to the source.

Abstract: Complex computer system architectures are described for utilizing a knowledge data graph comprised of elements, and selecting a discovery element to replace an existing element of a formulation depicted in the knowledge data graph. The substitution process takes advantage of the knowledge data graph structure to improve the computing capabilities of a computing device executing a substitution calculation by translating the knowledge data graph into an embedding space, and determining a discovery element from within the embedding space.

Abstract: Methods and systems for responding to changing conditions include training a model, using a processor, using trajectories that resulted in a positive outcome and trajectories that resulted in a negative outcome. Training is performed using an adversarial discriminator to train the model to generate trajectories that are similar to historical trajectories that resulted in a positive outcome, and using a cooperative discriminator to train the model to generate trajectories that are dissimilar to historical trajectories that resulted in a negative outcome. A dynamic response regime is generated using the trained model and environment information. A response to changing environment conditions is performed in accordance with the dynamic response regime.

Abstract: Field-programmable gate array and method to implement an artificial neural network. A trained model of the neural network is processed, in which weights are defined in a floating-point format, to quantize each set of weights to a respective reduced-precision format in dependence on effect of quantization on accuracy of the model. For each set of weights, a partitioning scheme is defined for a set of block memories of the apparatus such that a plurality k of those weights can be stored in each addressable location of the set of memories, wherein k differs for different sets of weights. The apparatus can be programmed to implement the neural network such that weights in each set are persistently stored in a set of block memories partitioned according to the partitioning scheme for that set of weights.

Abstract: The system and methods of the disclosed subject matter provide an experimentation framework to allow a user to perform machine learning experiments on tenant data within a multi-tenant database system. The system may provide an experimental interface to allow modification of machine learning algorithms, machine learning parameters, and tenant data fields. The user may be prohibited from viewing any of the tenant data or may be permitted to view only a portion of the tenant data. Upon generating an experimental model using the experimental interface, the user may view results comparing the performance of the experimental model with a current production model.

Abstract: A method for predicting a failure of a power control unit of a vehicle is provided. The method includes obtaining data from a plurality of sensors of the power control unit of a vehicle subject to simulated multi-load conditions, implementing a machine learning algorithm on the data to obtain machine learning data, obtaining new data from the plurality of sensors of power control unit of the vehicle subject to real multi-load conditions, implementing the machine learning algorithm on the new data to obtain test data, predicting a failure of the power control unit based on a comparison between the test data and the machine learning data.

Abstract: A method for multivariate time series prediction is provided. Each time series from among a batch of multiple driving time series and a target time series is decomposed into a raw component, a shape component, and a trend component. For each decomposed component, select a driving time series relevant thereto from the batch and obtain hidden features of the selected driving time series, by applying the batch to an input attention-based encoder of an Ensemble of Clustered dual-stage attention-based Recurrent Neural Networks (EC-DARNNS). Automatically cluster the hidden features in a hidden space using a temporal attention-based decoder of the EC-DARNNS. Each Clustered dual-stage attention-based RNN in the Ensemble is dedicated and applied to a respective one of the decomposed components. Predict a respective value of one or more future time steps for the target series based on respective prediction outputs for each of the decomposed components by the EC-DARNNS.